Skip to main content
← engineering

Luminal Raises $25M Series A to Scale Search Compiler and Inference Cloud

By Zero G Talent•

AI infrastructure startup Luminal has raised $25 million in a Series A round led by Standard Capital, with participation from seed investors Y Combinator, Felicis, SagaVC, and Liquid 2. First reported by Axios and detailed by Data Center Dynamics, the new capital supports the company's search-based machine learning compiler and inference serving cloud.

Founded in 2025 out of Y Combinator's S25 batch, Luminal previously raised a $5.3 million seed round led by Felicis Ventures, as reported by TechCrunch. The startup has grown its cloud footprint rapidly: after launching its inference cloud in late 2025, Luminal now processes roughly 90 trillion tokens per month and has partnered with Amazon Web Services to direct tokens to AWS Trainium chips.

"The cost of running AI shouldn’t limit who gets to build it," co-founder and CEO Joe Fioti told Data Center Dynamics. "We’re building the infrastructure to make high-performance inference more accessible, giving developers more choice beyond Nvidia's ecosystem without forcing them to compromise on speed or cost."

Squeezing Performance Out of Heterogeneous Silicon

Most machine learning deployments rely heavily on hand-optimized CUDA kernels and standard runtime stacks. Luminal tackles model execution from a different angle: an automated, search-based compiler written in Rust that discovers optimal GPU and accelerator kernels without requiring engineers to write them by hand.

The compiler translates PyTorch computations into a lightweight intermediate representation (IR), then relies on egglog and equality saturation rewrite rules (tiling, unrolling, vectorization, memory layout) to generate candidate kernels. The engine uses cost models and bandit-style search to explore trillions of candidate implementations and find the fastest valid kernel for a target chip. Luminal compiles and benchmarks these candidates directly on real hardware, verifying numerical correctness with property tests and equivalence checks before caching and reusing the best kernels across deployments.

In a lecture for GPU MODE, Fioti explained how Luminal executes full forward passes through "mega kernels." Instead of launching thousands of tiny kernels with separate launch overhead and boundary synchronization points that purge SRAM, Luminal uses an on-device interpreter on GPU cores. A global instruction queue coordinates cores using atomic operations and barrier synchronization, enabling dynamic work stealing, kernel fusion, and instruction prefetching across the forward pass.

In an interview hosted by Standard Capital, Fioti pointed out that hardware specialization makes software optimization increasingly critical. ASICs like Google's TPU or AWS Trainium reserve silicon area for raw compute, pushing memory prefetching and instruction scheduling onto compilers. By applying equality saturation to explore candidate schedules, Luminal bypasses static heuristics, allowing non-Nvidia hardware to run with throughput comparable to top-tier GPU setups.

The Engineering Team Behind the Architecture

Luminal was co-founded by Joe Fioti, Jake Stevens, and Matthew Gunton. Fioti previously worked on CPU microcode and ML accelerators at Intel, where he noticed that specialized hardware often sat idle because writing efficient software for it proved too difficult. He built the initial open-source version of Luminal to run local language models fast on MacBook hardware before expanding the runtime to CUDA.

Stevens, who serves as COO, previously worked at Apple focusing on image processing after studying imaging science at the Rochester Institute of Technology. Gunton, the company's CTO, was an engineer at Amazon where he built automated systems that identified and resolved issues across Amazon's fulfillment network.

While Luminal maintains its core compiler as open source, its primary revenue engine resembles Vercel's model: hosting and managing deployments through Luminal Cloud, paired with licensing its proprietary serving stack to enterprise customers running their own data centers.

Engineering and Growth Roles in San Francisco

Luminal is expanding beyond its initial core team of three, listing five open positions based on-site in downtown San Francisco. The roles target systems engineering, hardware-level optimization, and the company's commercial go-to-market motion.

On the compiler side, the team is hiring a Senior Compiler Engineer ($200,000 to $350,000) to build core compiler infrastructure in Rust. This work involves developing backend code generation for multiple hardware targets and implementing compiler passes for operator fusion and memory planning. The team is also seeking a Compiler Engineer ($150,000 to $250,000) to assist in core compiler design, write CUDA kernels, and run performance reviews on customer models.

To drive kernel efficiency on execution backends, the company is recruiting a Founding Kernel Engineer ($150,000 to $250,000). The role works directly with compiler engineers to investigate hardware bottlenecks, conduct benchmarking, and take low-level optimizations into production. At the platform level, the Cloud Inference Engineer ($150,000 to $250,000) will ship low-latency and high-throughput model serving on Luminal Cloud, tuning optimizations like KV caching and paged attention while advancing the platform's scheduler and autoscaler.

On the business side, Luminal is hiring its first sales lead, a Head of Growth ($175,000 to $200,000). Working directly with the founders, this hire will build the outbound sales motion from scratch, splitting focus between building pre-sales pipeline for the compiler and serving stack and managing post-sales adoption with high-volume inference customers.

Role Location Work Mode Salary
Head of Growth San Francisco, CA, US Onsite $175,000 – $200,000
Founding Kernel Engineer San Francisco, CA, US Onsite $150,000 – $250,000
Senior Compiler Engineer San Francisco, CA, US Onsite $200,000 – $350,000
Compiler Engineer San Francisco, CA, US Onsite $150,000 – $250,000
Cloud Inference Engineer San Francisco, CA, US Onsite $150,000 – $250,000

See all open roles at Luminal